JP5977275B2 - Branch recognition device - Google Patents

Description

  The present invention relates to a technique for accurately recognizing road branches.

Patent Document 1 describes a technique for detecting a lane having a lower parallelism as a lane branch, focusing on the parallelism with the estimated trajectory of the host vehicle.
Patent Document 2 describes a technique for determining branching by a change in lane width.

JP 2006-331389 A (page 17) JP 2000-105898 A (pages 14 to 16)

  However, in the technique described in Patent Document 1, parallelism and curvature greatly vary when it is difficult to estimate a road shape due to the presence of a plurality of lanes that are not branched roads, particularly white lines such as composite lines. In addition, it may be erroneously recognized as a branch road, and there is a problem that the estimation accuracy of the course shape is not good.

  In addition, the technique described in Patent Document 2 has a problem that a lane that is not a branch road is misrecognized as a branch road because the branch determination is performed with only one feature, and the estimation accuracy of the route shape is not good. It was.

  The present invention has been made in view of such problems, and an object thereof is to provide a technique for accurately recognizing a branch path.

  According to the branch recognition apparatus of the present invention, a white line candidate is extracted by image processing from an image around a vehicle acquired by an in-vehicle camera, a certainty factor of white line likelihood of the white line candidate is calculated, and based on the calculated certainty factor, Select the white line to be recognized, and the characteristics related to the selected white line are that the lane is not a compound line, that the lane is a solid line, that the lane is not a sharp curve, and that the parallelism of the left and right lanes varies from the set value. Is calculated, or the degree of variability in the curvature of the lane is greater than the set value. Since the presence / absence of the branch path is determined based on the certainty factor, the branch path can be recognized with high accuracy.

It is a functional block diagram showing the schematic structure of a branch path recognition apparatus. It is explanatory drawing showing the position of the vehicle-mounted camera in a vehicle. 2 is a functional block diagram illustrating a schematic functional configuration of the image processing apparatus. FIG. It is a flowchart which shows a branch path recognition process. (A) is explanatory drawing explaining a parallelism fluctuation | variation calculation process, (b) is explanatory drawing explaining a curvature fluctuation | variation calculation process, (c) is explanatory drawing explaining a composite linearity calculation process, (D) is explanatory drawing explaining a solid line / broken line likelihood calculation process, (e) is explanatory drawing explaining a sharp curve likeness calculation process.

Embodiments of the present invention will be described below with reference to the drawings. In addition, this invention is not limited to the following embodiment, It is possible to implement in various aspects.
[1. Description of the configuration of the branch path recognition apparatus 1]
As shown in FIG. 1, the branch path recognition device 1 includes an in-vehicle camera 10 that acquires an image around the vehicle, and an image processing device 20 that performs image processing on the image around the vehicle acquired by the in-vehicle camera 10. .

[1.1. Description of configuration of in-vehicle camera 10]
The in-vehicle camera 10 is composed of a CCD camera, and as shown in FIG. 2, the vehicle-mounted camera 10 is installed at the front center of the vehicle and continuously photographs the front thereof.

[1.2. Description of Configuration of Image Processing Device 20]
The image processing apparatus 20 is realized as a process executed on a well-known microcomputer including a CPU, ROM, RAM, EEPROM, DSP (digital signal processor), and the like. As shown in FIG. It functions as a candidate calculation unit 21, a white line feature calculation unit 22, a white line feature integration unit 23, a control target white line selection processing unit 24, and a branch path determination unit 25.

  The white line candidate calculation unit 21 has a function of calculating white line candidates on a road from an image acquired by the in-vehicle camera 10 by image processing. Specifically, from an image acquired by the in-vehicle camera 10, a white line-like thing on the road (white line candidate) based on the number of votes (solid line / broken line determination result) in the pattern matching and straight line extraction Hough transform processing that are known image processing ) Is calculated. A plurality of white line candidates may be extracted from one frame image.

The white line candidate calculation unit 21 corresponds to white line candidate extraction means.
The white line feature calculation unit 22 has a function of calculating the certainty of white line likelihood for the white line candidates extracted by the white line candidate calculation unit 21. The certainty factor is a numerical value (likelihood) set within a range of 0.01 to 1, and includes (1) line type (composite line) pattern determination processing and (2) solid line (voting number) determination processing. (3) Straightness determination processing, (4) Contrast strength determination processing, (5) Contrast saliency determination processing, (6) No pattern determination processing inside white line, (7) Distance determination processing from pedestrian crossing, ( 8) Set for each of the process of determining the height of luminance from the road surface, and the process of determining the distance from the target (9). Therefore, a plurality of certainty factors are set for one white line candidate.

  The white line feature integration unit 23 has a function of outputting a likelihood indicating the likelihood of a white line by multiplying and integrating the certainty calculated and output in each of the processes (1) to (9). The integration process in the white line feature integration unit 23 is determined from integration by multiplication of certainty factors based on a Bayesian estimation framework.

The white line feature calculation unit 22 and the white line feature integration unit 23 correspond to a certainty factor calculation unit.
The control target white line selection processing unit 24 has a function of selecting a white line candidate having the maximum likelihood among the likelihoods output by the white line feature integration unit 23 as a white line on the road. This control target white line selection processing unit 24 uses, as characteristics, that it is the innermost solid line for the own vehicle, that it is not the outer line at the time of branch road determination, and that the left-right white line pair is likely to be a lane. Since the white line feature integration unit 23 outputs the likelihood that the “white line likelihood” is integrated with respect to the white line candidate, the control target white line selection processing unit 24 outputs the white line having the maximum integrated likelihood. Is selected as the white line on the road.

The control target white line selection processing unit 24 corresponds to a white line selection unit.
The branch path determination unit 25 calculates a certainty factor of the branch path from the characteristic regarding the white line selected by the control target white line selection processing unit 24 by executing a branch path recognition process described later, and based on the calculated confidence factor. It is a DSP having a function of determining the presence or absence of a branch path. In this branch path determination unit 25, as a feature, (1) the lane is not a composite line, (2) the lane is a solid line, (3) the lane is not a sharp curve, (4) left and right lanes Any one of the above-mentioned features (1) to (5) using the fact that the change in parallelism is larger than the set value and (5) the amount of change in the curvature of the lane is larger than the set value. The likelihood (integrated likelihood) obtained by integrating the likelihood of the two or the likelihood of two or more features is calculated, and the presence / absence of a branch path is determined based on the calculated likelihood.

The branch path determination unit 25 corresponds to a branch path determination unit.
[2. Explanation of branch recognition processing]
Next, the branch path recognition process executed by the branch path determination unit 25 of the image processing apparatus 20 will be described with reference to the flowchart of FIG.

This process is repeatedly executed every predetermined time while the host vehicle is traveling.
First, in the first step S110, the parallelism is calculated. Specifically, the branch path determination unit 25 calculates the parallelism of the white line selected by the control target white line selection processing unit 24. Thereafter, the process proceeds to S120.

In S120, the curvature is calculated. Specifically, the branch path determination unit 25 calculates the curvature of the white line selected by the control target white line selection processing unit 24. Thereafter, the process proceeds to S130.
In S130, the likelihood of the parallelism fluctuation is calculated. Specifically, the branch path determination unit 25 calculates the variation of the parallelism of the left and right lanes calculated in S110, and calculates the likelihood of the calculated variation of the parallelism. More specifically, the likelihood of the change in parallelism is calculated with reference to the table shown in FIG. Here, the likelihood is set to 0.5 when the variation in parallelism between the left and right lanes is small, and the likelihood is set to 0.1 when the variation in parallelism between the left and right lanes is large. Note that the value of 0.5 is a value that is set in order to exhibit a function of suppressing branch determination when the variation in parallelism is small, and a value of 0.1 is strong for other multiple IPs. It is a value that is set to reverse branch suppression only when it is determined that the branch is a certainty. Thereafter, the process proceeds to S140.

  In S140, the likelihood of curvature variation is calculated. Specifically, the branch path determination unit 25 calculates the variation of the curvature of the white line calculated in S120, and calculates the likelihood of the calculated variation of the curvature. More specifically, the likelihood of the white line curvature fluctuation is calculated with reference to the table shown in FIG. Here, the likelihood is set to 0.5 when the curvature variation of the white line is smaller than the set value, and the likelihood is set to 0.1 when the curvature variation of the white line is larger than the set value. The value of 0.5 is a value set to exert a function of suppressing branch determination when the curvature variation of the white line is small, and the value of 0.1 is strong in other plural IPs. It is a value that is set to reverse branch suppression only when it is determined that the branch is a certainty. Thereafter, the process proceeds to S150.

  In S150, the likelihood of the composite line is calculated. Specifically, the branch path determination unit 25 calculates the likelihood of the composite line for the white line selected by the control target white line selection processing unit 24. More specifically, the likelihood of composite line likelihood is calculated with reference to the table shown in FIG. Here, the likelihood is set to 0.5 when the composite line likelihood is smaller than the set value, and the likelihood is set to 0.1 when the composite line likelihood is larger than the set value. Note that this value of 0.5 is a value that is set to exhibit the function of suppressing branch determination when the composite line likelihood is small, and a value of 0.1 is a strong confidence that other IPs are strong. This value is set to reverse branch suppression only when it is determined that a branch is taken. Thereafter, the process proceeds to S160.

  In S160, the likelihood of a solid line / broken line is calculated. Specifically, the branch path determination unit 25 calculates the likelihood that the white line selected by the control target white line selection processing unit 24 is a solid line or a broken line. More specifically, the likelihood of the likelihood of a broken line is calculated with reference to the table shown in FIG. Here, when the likelihood of a broken line is smaller than the set value, the likelihood is set to 0.5, and when the likelihood of a broken line is larger than the set value, the likelihood is set to 0.1. Note that this value of 0.5 is a value that is set to exhibit the function of suppressing branch determination when the likelihood of a broken line is small, and a value of 0.1 is a certainty that other IPs are strong. Only when it is determined to be a branch, this value is set to reverse branch suppression. Thereafter, the process proceeds to S170.

  In S170, the likelihood of road shape information (curve) is calculated. Specifically, the branch path determination unit 25 calculates the likelihood that the white line selected by the control target white line selection processing unit 24 is a sharp curve. More specifically, the likelihood of a sharp curve is calculated with reference to the table shown in FIG. Here, when the likelihood of a sharp curve is smaller than the set value, the likelihood is set to 0.5, and when the likelihood of a sharp curve is larger than the set value, the likelihood is set to 0.1. Note that this value of 0.5 is a value that is set in order to exhibit the function of suppressing branch determination when the likelihood of a sharp curve is small, and a value of 0.1 is certain that other multiple IPs are strong. This value is set to reverse branch suppression only when it is determined that a branch is taken. Thereafter, the process proceeds to S180.

  In S180, the integrated likelihood is calculated. Specifically, the branch path determination unit 25 calculates the likelihood of the parallelism fluctuation calculated in S130, the likelihood of the curvature fluctuation calculated in S140, the likelihood of the composite line calculated in S150, and the solid / broken line calculated in S160. The integrated likelihood is calculated by multiplying and integrating the likelihood of the sharp curve calculated in S170. Thereafter, the process proceeds to S190.

  In S190, it is determined whether the branch integration likelihood is a predetermined value or more. Specifically, the branch path determination unit 25 determines whether or not the integrated likelihood calculated in S180 is 50% or more as a predetermined value. When it is determined that the integrated likelihood is 50% or more (S190: YES), the process proceeds to S210. On the other hand, when it is determined that the integrated likelihood is less than 50% (S190: NO), the process proceeds to S200.

  In S200, normal detection is performed. Specifically, upon determining that the integrated likelihood is less than 50% in S190, the branch path determination unit 25 performs normal detection for detecting the white line candidate as a white line. Thereafter, the process proceeds to S220.

  In S210, white line candidates are excluded. Specifically, upon determining that the integrated likelihood is 50% or more in S190, the branch path determination unit 25 excludes the white line candidate. That is, it determines with it being a branch path. Thereafter, the process proceeds to S220.

  In S220, the detection result is output. Specifically, when the branch path determination unit 25 proceeds from S200 to S220, it outputs as a detection result that the white line candidate has been detected as a white line, whereas when it proceeds from S210 to S220, The fact that the white line candidate is excluded and it is determined as a branch road is output as a detection result. By this process, the branch path determination unit 25 corresponds to a determination result output unit. Thereafter, this process is terminated.

[3. Effects of the embodiment]
According to the branch road recognition apparatus 1 of the present embodiment, the branch road determination unit 25 is a feature related to the white line selected by the control target white line selection processing unit 24, and (1) the lane is not a compound line, (2 ) The lane is a solid line, (3) The lane is not a sharp curve, (4) The change in parallelism of the left and right lanes is greater than the set value, and (5) The amount of change in the lane curvature is the set value Is used, and the likelihood (integrated likelihood) obtained by integrating the likelihood of any one of the features (1) to (5) above or the likelihood of two or more features is calculated. Since the presence / absence of the branch path is determined based on the calculated likelihood, the branch path can be recognized with high accuracy.

  DESCRIPTION OF SYMBOLS 1 ... Branch road recognition apparatus, 10 ... Car-mounted camera, 20 ... Image processing apparatus, 21 ... White line candidate calculation part, 22 ... White line feature calculation part, 23 ... White line feature integration part, 24 ... Control target white line selection process part, 25 ... A branch path determination unit.

Claims (3)

White line candidate extraction means (21) for extracting a white line candidate by image processing from an image around the vehicle acquired by the in-vehicle camera (10);
A certainty factor calculating means (22, 23) for calculating a certainty factor of white line likeness of the white line candidate;
A white line selecting means (24) for selecting a white line to be recognized based on the certainty factor calculated by the certainty factor calculating means;
A branch path determination means (25) for determining the presence or absence of a branch path based on the characteristics relating to the white line selected by the white line selection means;
With
The branch road judging means (25) is characterized in that the lane is not a composite line, the lane is a solid line, the lane is not a sharp curve, Calculate the certainty factor that integrates the certainty factors of two or more of the following: lane parallelism fluctuation is greater than the set value, or lane curvature fluctuation amount is greater than the set value, A branch path recognition device (1) characterized by determining the presence or absence of a branch path based on the calculated certainty factor. In the branch recognition apparatus (1) according to claim 1,
The branch path determination means (25) determines that a branch path exists when the calculated certainty factor is greater than or equal to a predetermined value, and determines that there is no branch path when the calculated certainty factor is less than a predetermined value. A branching path recognition device (1) characterized by determining. In the branch path recognition device (1) according to claim 1 or 2,
A branch path recognition device (1) comprising: a determination result output means (25) for outputting a determination result by the branch path determination means (25).